J. A. Guerrero, J. Matkowski, N. Merentes
Uniformly continuous composition operators in the space of functions of two variables of bounded
ϕ-variation in the sense of Wiener
Abstract. Assume that the generator of a Nemytskii composition operator is a function of three variables: the first two real and third in a closed convex subset of a normed space, with values in a real Banach space. We prove that if this operator maps a certain subset of the Banach space of functions of two real variables of boun- ded Wiener ϕ-variation into another Banach space of a similar type, and is uniformly continuous, then the one-sided regularizations of the generator are affine in the third variable.
2000 Mathematics Subject Classification: 47B33, 26B30, 26B40.
Key words and phrases: ϕ-variation in the Wiener sense, composition operator, uni- formly continuous operator, left-left regularization, Jensen equation.
1. Introduction. Let Iabdenote the rectangle [a1, b1]×[a2, b2] with the vertices a = (a1, a2), b = (b1, b2) ∈ ℝ2. Let (X, | · |), (Y, | · |) be real normed spaces and C be a set in X. For a function h : Iab× C −→ Y define the Nemytskii composition operator H : CIab−→ YIab by
H(f )(t, s) = h(t, s, f (t, s)), f ∈ CIab, (t, s)∈ Iab,
where CIab stands for the family of all functions f : Iab −→ C. The function h is called the generator of the composition operator H.
Let (BVϕ(Iab, X),k · kϕ) be the Banach space of functions f ∈ XIab which are of bounded ϕ-variation in the sense of Wiener [9], where the norm k · kϕ is defined with the aid of Luxemburg-Nakano-Orlicz seminorm [4, 7, 8].
Assume that H maps the set of functions f ∈ BVϕ(Iab, X) such that f (Iab) ⊂ C into BVψ(Iab, Y ). In the present paper we prove that, under some conditions, if H is uniformly continuous, then the left-left, right-right, left-right and right-left
regularizations of its generator h with respect to the two first variables are affine functions in the third variable.
This generalizes the result of Chistyakov [2] where it is assumed that H is Lipschitzian.
2. Preliminaries. We begin this section with some notations.
Let a = (a1, a2), b = (b1, b2) ∈ ℝ2 be such that ai< bi, i = 1, 2. In the sequel Iab := [a1, b1] × [a2, b2] denotes the basic rectangle. Let ξ = (ti)mi=0 and η = (sj)nj=0 be partitions of [a1, b1] and [a2, b2], respectively (i.e., m, n ∈ ℕ, a1 = t0 < t1 <
· · · < tm= b1 and a2 = s0< s1 <· · · < sn = b2). For each function f ∈ XIab, we define
∆10f (ti, sj) = f(ti, sj) − f(ti−1, sj)
∆01f (ti, sj) = f(ti, sj) − f(ti, sj−1)
∆11f (ti, sj) = f(ti−1, sj−1) − f(ti−1, sj) − f(ti, sj−1) + f(ti, sj).
We denote by F the set of all non-decreasing continuous functions ϕ : [0, +∞) −→
[0, +∞) such that ϕ(t) = 0 if and only if t = 0, and limt−→∞ϕ(t) =∞.
Definition 2.1 (Chistyakov [2])
Let ϕ ∈ F, X be a real normed space and f ∈ XIab be a function.
(i) For x2 ∈ [a2, b2], the Jordan ϕ-variation of the function f(·, x2) in [x1, y1] ⊂ [a1, b1], denoted vϕ,[x1,y1](f) = vϕ,[x1,y1](f(·, x2)), is defined by
(1) vϕ,[x1,y1](f) := sup
ξ
Xm i=1
ϕ (|∆10f (ti, x2)|) ,
where the supremum is taken over all the partitions ξ = (ti)mi=0 of [x1, y1].
(ii) For x1 ∈ [a1, b1], the Jordan ϕ-variation of the function f(x1,·) in [x2, y2] ⊂ [a2, b2] is defined by
(2) vϕ,[x2,y2](f) := sup
η
Xn j=1
ϕ (|∆01f (x1, sj)|) ,
where the supremum is taken over all the partitions η = (sj)nj=0 of [x2, y2].
(iii) The (two-dimensional) Hardy-Vitali-Wiener ϕ-variation of a function f ∈ XIab, denoted vϕ,Iab(f), is defined by
(3) vϕ,Iab(f) := sup
(ξ,η)
Xm i=1
Xn j=1
ϕ (|∆11f (ti, sj)|) ,
where the supremum is taken over all pairs m, n ∈ ℕ and (ξ, η) with ξ a partition of [a1, b1] and η a partition of [a2, b2] of the above form.
(iv) The Wiener total ϕ-variation of f ∈ XIab is defined by
(4) T Vϕ(f) = T Vϕ(f, Iab) := vϕ,[a1,b1](f(·, a2)) + vϕ,[a2,b2](f(a1,·)) + vϕ,Iab(f).
(v) A function f ∈ XIab is of bounded Wiener total ϕ-variation in Iab, if T Vϕ(f) <
∞.
(vi) The class of all functions f ∈ XIba with finite Wiener total ϕ-variation is denoted by Vϕ(Iab, X); that is
Vϕ(Iab, X) :=n
f ∈ XIab : T Vϕ(f) < ∞o .
We denote by BVϕ(Iab, X) the vector space generated by Vϕ(Iab, X), i.e.
BVϕ(Iab, X) =n
f ∈ XIab : ∃ λ > 0 such that λf ∈ Vϕ(Iab, X)o . In the space BVϕ(Iab, X) we define the norm
kfkϕ:= |f(a)| + pϕ(f), where
pϕ(f) = inf { > 0 : T Vϕ(f/) ¬ 1} . Some properties of pϕare in the following lemma.
Lemma 2.2 (Chistyakov [2]) For f ∈ BVϕ(Iab, X), we have (a) if (t, s), (t0, s0) ∈ Iab, then |f(t, s) − f(t0, s0)| ¬ 4ϕ−1(1)pϕ(f);
(b) if pϕ(f) > 0 then T Vϕ
f
pϕ(f)
¬ 1;
(c) if λ > 0 then
(c1) pϕ(f) ¬ λ if and only if T Vϕ
f λ
¬ 1;
(c2) if T Vϕ(f/λ) = 1 then pϕ(f) = λ.
Theorem 2.3 (Chistyakov [2]) If ϕ ∈ F is convex and X is a Banach space, then BVϕ(Iab), k · kϕis a Banach space.
In the sequel we will use the notions of left-left regularization and the left- left continuity of a function of two variables. Let f : Iab −→ X. If the function f− : Iab−→ X given by
f−(x1, x2) :=
limy1→x− 1
y2→x−2
f (y1, y2) for (x1, x2) ∈ (a1, b1] × (a2, b2], limy1→x−
1
y2→a+2
f (y1, y2) for x1∈ (a1, b1] and x2= a2, limy1→a+
1
y2→x−2
f (y1, y2) for x1= a1 and x2∈ (a2, b2], limy1→a+
1
y2→a+2
f (y1, y2) for x1= a1 and x2= a2.
is well defined, that is, if all the above limits exist, then f− is called the left-left regularization of f.
Remark 2.4
It is to be noted that (y1, y2) → (x−1, x−2) means that (y1, y2) ∈ Iab, yi< xi, i = 1, 2, and (y1, y2) → (x1, x2) inℝ2, and similarly for the other limits.
Remark 2.5
In a similar way we can define the right-right, left-right, right-left regularizations of a function f ∈ BVϕ(Iab, X).
Definition 2.6 A function f : Iab−→ X is said to be left-left continuous if
(x,y)→(tlim−,s−)f (x, y) = f (t, s) for all (t, s) ∈ (a1, b1] × (a2, b2].
Lemma 2.7 (Chistyakov [2]) If f ∈ BVϕ(Iab, X) then f−exists, f− ∈ BVϕ−(Iab, X) and f− is left-left continuous.
Remark 2.8
The respective counterparts of Lemma 2.7 for the right-right, left-right and right-left regularizations hold also true.
We denote by BVϕ−(Iab, X) the subspace of BVϕ(Iab, X) of those functions which are left-left continuous on (a1, b1] × (a2, b2].
3. Main result. Denote by A(X, Y ) the space of all additive mappings A : X −→ Y and by L(X, Y ) the space of all continuous linear mappings A : X −→ Y .
The main result of this section reads as follows:
Theorem 3.1 Let Iab ⊂ ℝ2 be rectangle, (X, | · |) be a real normed space, (Y, | · |) be a real Banach space, C be a closed convex subset in X and assume that ϕ, ψ ∈ F. If the composition operator H generated by h : Iab× C −→ Y maps BVϕ(Iab, C) into BVψ(Iab, Y ), and is uniformly continuous, then there exist functions A : Iab−→ A(X, Y ) and B : Iab−→ Y such that
h−(t, s, x) = A(t, s)x + B(t, s), (t, s) ∈ Iab, x∈ C,
where h− is the left-left regularization of the function (t, s) → h(t, s, x). Moreover, if 0 ∈ C and intC 6= ∅, then A : Iab−→ L(X, Y ) and B ∈ BVψ−(Iab, Y ).
Proof For every x ∈ C, the constant function Iab 3 (t, s) −→ x belongs to BVϕ(Iab, C). Since H maps BVϕ(Iab, C) into BVψ(Iab, Y ), the function Iab 3 (t, s) 7→
h(t, s, x) belongs to BVψ(Iab, Y ). Now Lemma 2.7 implies the existence of the left- left regularization h− of function (t, s) → h(t, s, x). By assumption, H is uniformly continuous on BVϕ(Iab, C). Let ω : ℝ+ −→ ℝ+ be the modulus continuity of H, that is
ω(ρ) := sup
kH(f1) − H(f2)kψ: kf1− f2kϕ¬ ρ; f1, f2∈ BVϕ(Iab, C)
, for ρ > 0.
Hence we get
(5) kH(f1) − H(f2)kψ ¬ ω (kf1− f2kϕ) , for f1, f2∈ BVϕ(Iab, C).
From the definition of the norm k · kψ we obtain
(6) pψ(H(f1) − H(f2)) ¬ kH(f1) − H(f2)kψ, for f1, f2∈ BVϕ(Iab, C).
From Lemma 2.2(c1) and (6), if ω (kf1− f2kϕ) > 0, then (7) vψ,[a1,b1]
(H(f1) − H(f2)) (·, a2) ω (kf1− f2kϕ)
¬ T Vψ
H(f1) − H(f2) ω (kf1− f2kϕ)
¬ 1.
Therefore, for any a1 = α1 < β1 < α2 < β2 <· · · < αm< βm= b1; a2 = α1 <
β1< α2 < β2<· · · < αm< βm = b2, m∈ ℕ, the definitions of the operator H and the functional vψ,[a1,b1](·, a2), imply that
(8)
Pm
i=1ψ|h(α
i,αi,f1(αi,αi))−h(αi,αi,f2(αi,αi))−h(αi,βi,f1(αi,βi))+h(αi,βi,f2(αi,βi)) ω(kf1−f2kϕ)
−h(βi,αi,f1(βi,αi))+h(βi,αi,f2(βi,αi))+h(βi,βi,f1(βi,βi))−h(βi,βi,f2(βi,βi))|
ω(kf1−f2kϕ)
¬ 1.
For α, β ∈ ℝ, α < β, we define functions ηα,β :ℝ −→ [0, 1] by
(9) ηα,β(t) :=
0 if t ¬ α
t−α
β−α if α ¬ t ¬ β 1 if β ¬ t .
We first fix t ∈ (a1, b1], s ∈ (a2, b2], m ∈ ℕ. For arbitrary finite sequence a1< α1<
β1< α2< β2<· · · < αm< βm< t, a2< α1< β1< α2< β2<· · · < αm< βm<
s and x1, x2∈ C, x16= x2, the functions f1, f2: Iab−→ X defined by (10)
f`(τ, γ) := 1 2
hηαi,βi(τ) + ηα
i,βi(γ) − 1
(x1− x2) + x`+ x2
i, (τ, γ)∈ Iab, ` = 1, 2,
belong to BVϕ(Iab, C). From (10) we have
f1− f2=x1− x2
2 ,
therefore
kf1− f2kϕ= x1− x2
2 ; and, moreover,
f1(αi, ¯αi) = x2; f2(αi, ¯αi) = −x1+ 3x2
2 ; f1(βi, ¯αi) = x2; f2(βi, ¯αi) = −x1+ 3x2
2 ;
f1(βi, ¯βi) = x1; f2(βi, ¯βi) = x1+ x2
2 ; f1(αi, ¯βi) =x1+ x2
2 ; f2(αi, ¯βi) = x2. Using (8), we hence get
(11)
Xm i=1
ψ
h(αi, αi, x2) − h(αi, αi,−x1+3x2 2) − h(αi, βi,x1+x2 2) + h(αi, βi, x2) ω(kf1− f2kϕ)
−h(βi, αi, x2) + h(βi, αi,−x1+3x2 2) + h(βi, βi, x1) − h(βi, βi,x1+x2 2) ω(kf1− f2kϕ)
!
¬ 1.
Since, for any x ∈ C, the constant function Iab3 (t, s) −→ x belongs to BVϕ(Iab, C) and H maps BVϕ(Iab, C) into BVψ(Iab, Y ), the function Iab 3 (t, s) 7→ h(t, s, x) is in BVψ(Iab, Y ) for any fixed x∈ C. From continuity of ψ and the left-left continuity of h− (Lemma 2.7), letting (α1, α1) tend to (t, s) from the left in (11), we obtain
Xm i=1
ψ
h−(t, s, x1) − 2h− t, s,x1+x2 2+ h−(t, s, x2) ω|x
1−x2| 2
¬ 1,
that is ψ
h−(t, s, x1) − 2h− t, s,x1+x2 2+ h−(t, s, x2) ω|x
1−x2| 2
¬ 1 m. Hence, since m ∈ ℕ is arbitrary,
ψ
h−(t, s, x1) − 2h− t, s,x1+x2 2+ h−(t, s, x2) ω
|x1−x2| 2
= 0.
As ψ ∈ F, we obtain
h−(t, s, x1) − 2h−
t, s,x1+ x2 2
+ h−(t, s, x2) = 0.
Therefore
(12) h−
t, s,x1+ x2
2
= h−(t, s, x1) + h−(t, s, x2) 2
for all (t, s) ∈ (a1, b1] × (a2, b2], and for all x1, x2∈ C.
For t ∈ (a1, b1] and s = a2, let us fix a1 < α1 < β1 < α2 < β2 <· · · < αm <
βm< t and a2< α1< β1 < α2< β2<· · · < αm< βm< b2. Proceeding as above we get (11). Taking the limit as (α1, βm) → (t−, a+2) and by (11), we get, again (12). The cases when t = a1 and s ∈ (a2, b2] or t = a1 and s = a2 can be treated similarly. Consequently,
h−
t, s,x1+ x2
2
= h−(t, s, x1) + h−(t, s, x2) 2
is valid for all (t, s) ∈ Iab and all x1, x2∈ C.
Therefore, the function h−(t, s, ·) satisfies the Jensen functional equation in C, for each (t, s) ∈ Iab. Adapting the standard argument (cf. Kuczma [3]), we conclude that, for each (t, s) ∈ Iab there exist an additive function A(t, s) and B(t, s) ∈ Y such that
(13) h−(t, s, x) = A(t, s)x + B(t, s), x∈ C, (t, s) ∈ Iab.
The “moreover part”, follow from the uniform continuity of the operator H : BVϕ(Iab, C) −→ BVψ(Iab, Y ) and intC 6= ∅ imply the continuity of the func- tion A(t, s), consequently A(t, s) ∈ L(X, Y ).
Since 0 ∈ C, putting x = 0 in (13), we get
h−(t, s, 0) = B(t, s), (t, s) ∈ Iab,
which shows that B ∈ BVψ(Iab, Y ). ■
Remark 3.2 The counterparts of Theorem 3.1 for the right-right, right-left and left-right regularizations are also valid.
Remark 3.3 The uniformly continuous composition operators for functions of bounded variation in a single variable were considered in [5].
References
[1] J. Appell and P. P. Zabrejko, Nonlinear Superposition Operator, Cambridge University Press, New York, 1990.
[2] V. V. Chistyakov, Superposition Operators in the Algebra of Functions of two Variables with Finite Total Variation, Monatshefte f¨ur Mathematik 137 (2002), 99-114.
[3] M. Kuczma, An Introduction to the Theory of Functional Equations and Inequalities, Polish Scientific Editors and Silesian University, Warszawa -Kraków- Katowice, 1985.
[4] W. A. Luxemburg, Banach Function Spaces, Ph.D. thesis, Technische Hogeschool te Delft, Netherlands, 1955.
[5] J. Matkowski, Uniformly continuous superposition operators in the space of bounded variation functions, Math. Nachr.282 (2010).
[6] J. Matkowski and J. Miś, On a Characterization of Lipschitzian Operators of Substitution in the Space BV ([a, b]), Math. Nachr.117 (1984), 155-159.
[7] H. Nakano, Modulared Semi-Ordered Spaces, Tokyo, 1950.
[8] W. Orlicz, A note on modular spaces. I, Bull. Acad. Polon. Sci. Ser. Sci. Math. Astronom.
Phys.9 (1961), 157-162.
[9] N. Wiener, The quadratic variation of function and its Fourier coefficients, Massachusetts J.
Math.3 (1924), 72-94.
J. A. Guerrero
Universidad Nacional Experimental del T´achira, Dpto. de Matem´aticas y F´ısica San Cristóbal-Venezuela
E-mail: jaguerrero4@gmail.com, jguerre@unet.edu.ve J. Matkowski
Faculty of Mathematics, Computer Science and Econometrics, University of Zielona Góra Zielona Góra, Poland
Institute of Mathematics, Silesian University Katowice, Poland
E-mail: J.Matkowski@wmie.uz.zgora.pl N. Merentes
Universidad Central de Venezuela, Escuela de Matem´aticas Caracas-Venezuela
E-mail: nmer@ciens.ucv.ve
(Received: 28.09.2009)